Walk into any stockroom on count day and you'll hear the same conversation. The sheet says forty-one bottles, the shelf says thirty-six, and nobody can explain the other five. So someone types thirty-six into the system, the gap disappears from view, and the business moves on. Until next week, when it's back.
Most teams treat the gap as weather: unpleasant, unavoidable, not worth investigating. It isn't weather. Stock drift has a short list of causes, and once you can see which one you're paying for, each has a boring, practical fix.
Where the drift comes from
The first cause is consumption that never touches the till. Breakage, spillage, staff drinks, the plate sent back and remade: all real stock leaving the building with no transaction behind it. If there's no thirty-second way to record 'broke two glasses, dropped a bottle', it doesn't get recorded, and the shelf quietly diverges from the sheet.
The second is receiving. When a delivery arrives during service, the fastest move is to receive the purchase order exactly as ordered and deal with reality later. But the supplier shorted you a case, substituted a size, or the driver miscounted, and now the system believes stock you never had.
The third is units. You buy gin by the bottle and sell it by the pour. If the conversion lives in someone's head rather than in the system, every till reading drains stock in units that don't match how you count it.
The fourth is timing. Counts taken while the shop trades are chasing a moving target: every sale between the first shelf and the last one is drift you created yourself.
Count less, count better
The instinct is to count everything more often. The better move is to count less stock, more deliberately.
- Cycle-count by velocity: your twenty fastest movers weekly, the middle monthly, the dust quarterly.
- Count blind. If the counter can see the expected number, they'll find it. Enter what's on the shelf, then compare.
- Record a reason with every adjustment. 'Breakage', 'wastage', 'supplier short'. The categories are the diagnosis.
- Count at close or before open, never mid-trade.
Do this for six weeks and the variance report stops being noise and starts being a map. One venue we work with discovered eighty percent of their variance sat in four SKUs, all of them behind one bar, all of them poured, none of them measured.
What good looks like
Well-run operations don't have zero variance; they have explained variance. The count matches the sheet to within a percent, the exceptions carry reasons, and the Sunday ritual takes twenty-five minutes instead of four hours, because you're confirming a number you already trust, not reconstructing one you don't.
Stocked is the system these notes are written from. Start free or read more from the blog.